Mitigating demographic disparities in explanation fairness for hate speech moderation
In brief
- The study introduces Plausibility-EFG to measure explanation disparities across demographic groups.
- Adversarial training reduces explanation gaps but does not eliminate predictive biases.
- The Fair-REF framework improves alignment between automated explanations and human rationales.
This study examines fairness in explanations generated by hate speech detection models. Results show adversarial training reduces disparities between demographic groups without fully correcting predictive biases. A new framework, Fair-REF, is proposed to better align automated explanations with human rationales.
What it changes for you
- LegalHate speech moderation models must include explanation fairness metrics to prevent biased decisions against specific demographic groups.
This summary gives you the essentials. The full article, its figures and quotes are on cs.AI updates on arXiv.org.
Read the original articleAI-generated summary of the source, checked by rules. Report an error